RAINFALL PREDICTION FOR MINNA METROPOLIS USING ARTIFICIAL NEURAL NETWORK

ABSTRACT

The effect of rainfall in our society today is stupendous.  Rainfall is seen as a benefit to crops and lives. Accurate and timely rainfall prediction can be very helpful for effective security measures  for planning water resources  management,  transportation  activities, agricultural  tasks,  managing  flights  operations,  issuance  of flood warning  and flood situation.  This study aims to predict the rainfall of Minna metropolis.  Atmospheric data comprising  those  of maximum  temperature,  minimum  temperature,  relative  humidity and rainfall for four consecutive years spanning from January 2015 –  December 2018 were acquired from the Geography Department  of Federal University  of Technology, Minna.  The datasets were preprocessed and normalised,  and further partition into three parts:  70% for training set,  15% for testing set and 15% for validating set. Feed forward neural network  and binary classification  was used for the prediction.  The target data (rainfall) was labelled as positive or negative (rainfall or no rainfall),  that is, (1  or 0) with threshold  of 0.5 for classifying  the rainfalls.  The outcomes  of prediction  were evaluated  using  confusion  matrix.  The  best  test  result  indicates  that  66  days  were predicted  to have rainfall and  120  days predicted  for no rainfall with 69% accuracy, 1.3% error, 63% sensitivity and 84% specificity.  The best validated results also indicate that 77 days were predicted to have rainfall and 109 days predicted for no rainfall with 59% accuracy,  1.4% error, 52% sensitivity and 78% specificity.  The performance of the classifier is 0.568 (AUC = 57%).

CHAPTER  ONE

1.0 INTRODUTION

1.1 Background  to the Study

Rainfall is a natural phenomenon whose prediction is challenging and demanding as the world continues to witness  an ever changing climate conditions. Its forecast plays an important role in water resource management and therefore, it is of particular relevance to the agricultural sector,  which contributes significantly to the economy of any nation Abdulkadir et al. (2012). Rain in Nigeria increases from the coastal region, with annual rainfall  greater  than  3500 mm,  to the  Sahel region  in the north-western  and north• eastern parts,  with annual rainfall less than 600 mm (Omonona and Akintunde,  2009). The inter-annual variability of rainfall, particularly in the northern parts often results in climate hazards, especially  floods and erosion with their devastating  effects  on farm products and associated calamities and sufferings.

Several neighbourhoods  in Minna,  the capital  of Niger  state in the north  central  of Nigeria are under the threat of flash flood from heavy rainfall.  The residents of Dusten Kura area in Minna woke up to a big shock of rain water flooding their homes after a heavy rainfall on 13″ July 2017.  The whole incident affected the movement of vehicles and people due to water overflowing the drainage canal. On 2″ September 2012, two children were drowned while trying to cross a flooded drain caused by heavy rainfall in Minna  (NSEMA,  2012).  According  to Niger  State Emergency  Management  Agency (NSEMA), at least fourteen persons died due to flooding in different parts of Minna and over sixty houses were affected by flood due to heavy rainfall in Kontagora,  Tafa and Suleja Local Government Areas of the State (NSEMA,  2018).  On 9 September 2018, according to The Guardian Newspaper,  over thirty villages were affected  by flood in Mokwa Local Government Area of the State.  Houses,  farmlands and live stocks were destroyed  in this disaster  (Babalola, 2014). In most cases, floods were associated  with abnormally high daily rainfall events (Urnar, 2012).

The effect  of rainfall  on human  civilisation  is colossal. Rainfall  means crops;  and crop means  life.  Additionally,  rainfall  has a strong  influence  on the operation  of darns and reservoirs,  sewage  systems,  traffic  and  other  human  activities.  Previous  studies  have shown that among the entire climate  elements,  rainfall  is the most variable  element  in Nigeria  both temporally  and spatially  which  can have  significant  impact  on economic activities  (Kowal and Kanabe,  1972;  Kowal  and Kassam,  1978).  Rainfall  is one of the challenging  tasks in weather  forecasting.  Weather  data consists of various  atmospheric features such as wind,  precipitation,  humidity,  pressure,  and temperature  among others. Accurate   and  timely  rainfall  prediction   can  be  very  helpful   for  effective   security measures  for planning  water  resources  management,  issuance  of early  flood warning, construction  activities,  transportation  activities,  agricultural  tasks,  managing  the  flight operations  and flood  situation.  Data mining  techniques  can effectively  predict  rainfall by extracting  the hidden patterns  among available  features of past weather  data (Aftab and Ahmad,  2018).  The variability of rainfall is a crucial phenomenon  in today’s world. It is ever challenging and a topic of interest because prediction is not always accurate.  It is a continuous, high dimensional, dynamic and complicated process because it involves many  factors  of the atmosphere.  The parameters  required  to predict  the weather  are enormously complex such that there is uncertainty in prediction even for a short period (Geetha and Nasira, 2014).

In Nigeria and on the worldwide  scale,  large numbers of attempts have been made by different researchers  to predict rainfall accurately using various techniques,  but due to the nonlinear nature of rainfall, prediction accuracy obtained by these techniques is still below the satisfactory level. Of course, as with anything else, too much rain can lead to a host of problems.  Heavy rainfall can lead to numerous hazards, for example:  flooding, including risk to human  life, loss of crops and livestock, landslides which can threaten human  life,  disrupt transport  and communications,  and cause  damage  to building  and infrastructure. The increase  in rainfall will also improve water  availability, a condition which  will  impact  on  water  supply  and  improve  sanitation  and  health  care  delivery (Ifabiyi  and Ashaolu,  2013).  Therefore,  it is important  to evaluate  how rainfall  varies and how  it will be in the future to minimise  and reduce  the negative  impact  of heavy rainfall and to increase the society resilience to hazards such as floods and erosions. To achieve  this,  researchers   are  developing   and  applying  improved  weather  prediction models capable of accurately forecasting several events in Nigeria.

Artificial   Neural  Network   algorithm   becomes   an  attractive   inductive   approach   in rainfall  prediction  owing  to the non-linearity,  flexibility  and data learning  in building the   models   without   any  prior   knowledge   about   catchment   behaviours   and   flow processes.  In machine learning,  classification  can be referred to as task that requires the use of machine  learning  algorithms  that learn how to assign  a class label to examples from the problem  domain. Machine  learning  is a field of study and is concerned  with algorithms  that  learn  from  examples.  Classification   refers  to  a  predictive  modeling problem  where  a class  label  is predicted  for a  given  example  of input  data.  From  a modeling  perspective, classification  requires  a training  dataset with many examples  of input  and  output  from  which  to  learn.  A  model  will  use  the  training  dataset  and calculate how to best map examples  of input data to specific class labels.   Data mining algorithms  are  classified  as  supervised  and  un-supervised.  Supervised  methods   get trained first with pre-classified  data (training data) and then classify  the input data (test data)  (Ahmad  and  Aftab,  2017).  Un-supervised   methods  on  the  other  hand  do  not require  any training;  instead  of pre-classified  data,  these techniques  use  algorithms  to extract hidden  structure  from unlabeled  data.  It has been observed from latest research that for high accuracy, researchers prefer the integrated techniques for the rainfall prediction. In general, climate and rainfall are highly non-linear and complicated phenomena,  which  require  advanced  computer  modeling  and  simulation  for  their accurate prediction. An Artificial  Neural Network  (ANN)  can be used to predict  the behaviour of such nonlinear systems (Nayak, 2013).

1.2 Statement of the Research Problem

Heavy rainfall can lead to numerous destructions, for example; flooding, landslides and erosion which can threaten human life. It can also damage buildings and infrastructure, disrupt  transportation   and  communications   and  cause  losses  to  farm  crops  of the affected  areas.  Heavy rainfall has caused lots of damages and destruction to lives and properties in some parts of Minna, Niger State capital.

According  to  British  Broadcasting   Corporation   on  27  September  2018,  NEMA declared  state of emergency  in four states (Niger, Kogi, Anambra  and Delta)  due to destruction  by heavy  rainfall  (BBC New,  2018).  Excess  rain  brings  other  negative effects on the environment and even the economy of the affected location.  Existing rain prediction   are  mostly  numerical   and  traditional   in  nature   and  are  not  accurate (Gugulethu, 2013).

1.3 Aim and objectives of the study

The aim of this  study is to predict  the rainfall  of Minna metropolis  using Artificial Neural Network. The aim shall be achieved through the following objectives which are to:

1.        carry out interpolation of missing data and min-max normalisation technique for data scaling; and

11.          predict  the  number  of rainfall  days  usmg  classification  method  of Artificial

Neural Network as a predictive tool.

1.4 Study Area

Minna  is  the  headquarters  of Chanchaga  Local  Government  Area  of Niger  State, Nigeria.  It is the capital city of the state.  It lies between Latitudes 09°40′ 7.63″ N and 09° 39′ 59.72″ N and Longitudes 06° 30′ 0.32″ E and 06° 36′ 34.05″ E.  Figure 1.1  (a) and (b) is the map of the study area.

Minna lies on a valley bed (that is, lowland) bordered to the east by Paida hill stretching eastwards towards Maitumbi and bordered by Wushishi and Gbako to the West,  Shiroro to the North, Paikoro to the East and Katcha to the South.

Minna possesses  the tropical  continental  wet  and dry climate  based  on the Koppen Classification  Scheme and is characterised with two distinct seasons namely;  the wet season which begins  around March and runs through  October and dry season which begins from October to March.  The city has a mean annual rainfall of 1334 mm with September recording the highest rain of close to 330 mm on the average, while the least amount of rainfall occurs in December and January which can be as low as 1mm. Minna

1.5 Significance of the Study

A research of this nature is very important particularly to Minna residents,  Niger State Government  authorities  and the research  community  as it will enhance the safety of lives and properties  from rainfall hazards  due to better  awareness, preparedness  and planning by farmers,  aviation sector,  construction firms and disaster managers.  Also,  it will help to make rainfall prediction data available to all stakeholders.

1.6 Justification of the Study

The change in rainfall has implications in various sectors of the economy of Niger state. There is an increase in decade anomaly of rainfall in Minna (Akinsanola and Ogunjobi, 2014). According to Daramola et al. (2017), there are more wet years in the South and middle Belt ofNigeria which are prone to the occurrence of flooding.

On the 25″  to 26  August,  2014,  heavy downpour  spoiled most parts of Minna,  the Niger State capital, causing serious damages. It was gathered that houses, fences, mini bridges were washed away by the heavy rain.  Some of the affected areas were Barikin Sale and Farm centre in Tunga.  Others are Niteco, Nykangbe  and Kpankungu  areas (Babalola, 2014).

The areas that are previously affected by heavy rains and are still prone to flooding in Minna metropolis are Fadikpe, Barikin Sale, Shango. These areas are further shown in Figure 1.2

Hence,  this study is necessary based on the flooding history of Minna.  A prediction of heavy or low rainfall serves as an alarm to individual, communities and relevant government agencies.

1.7 Scope and Limitation  of the Study

This research  studied the rainfall variations  and classification  method was applied to predict  number  of rainfall  and  no  rainfall  days  in  Minna  using  four-year  dataset obtained  for  maximum  temperature,  minimum  temperature,  relative  humidity  and rainfall spanning from January 2015 – December 2018.

However, this study will be limited to rainfall prediction usmg four atmospheric parameters.

₦8,000.00 – DOWNLOAD FULL PROJECT DOWNLOAD FULL PROJECT Added to cart

---

---

---

---

Related Project Topics :